Veo 3.1 “Meta-Prompt” Builder for Creators: One Reusable Prompt That Generates Shot Cards, Variations, and Fixes (as of 2026-04-16)

If you’re a solo creator or a small marketing team, you don’t need more one-off prompt templates—you need one reusable meta-prompt you can paste into Gemini (or any LLM) to reliably generate:

Veo3Gen-ready shot cards in a consistent schema
Controlled variations (change one variable without derailing the whole concept)
Guardrails (must/avoid, safety-ish constraints, continuity)
Audio/dialogue direction
A fix list that helps you iterate fast without prompt chaos

This post walks you through building that meta-prompt, then using it as a repeatable workflow.

What a “meta-prompt” is (and when it beats writing prompts by hand)

A meta-prompt is a prompt that writes prompts—specifically, it instructs an LLM to output Veo-style prompts in a consistent structure.

Why it’s useful:

Consistency at scale: You get the same fields every time (subject, action, camera, lighting, style), which matters because Veo prompting benefits from specifying camera framing/motion, lighting, and style explicitly. (https://deepmind.google/models/veo/prompt-guide/)
Less creative drift: Instead of re-inventing your format for every concept, you keep a single “house style” and swap inputs.
Faster iteration: You can ask for “fix ONLY camera move” or “change ONLY setting” without rewriting everything.

As of 2026-04-16, Veo 3.1 is described as stable and generally available for production on Vertex AI, with professional-grade creative controls, multiple aspect ratios, and rich synchronous audio—so having a structured prompt workflow is worth it. (https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1)

The 5 outputs your meta-prompt should generate every time

1) Shot Cards (your core output)

Each shot card should be predictable and scannable. Veo’s own guidance emphasizes clarity around camera framing/motion, lighting, style, character detail, location detail, and action. (https://deepmind.google/models/veo/prompt-guide/)

In this workflow, every shot card uses a strict 6-line schema:

Subject
Action
Scene
Camera
Light
Style

2) Variations (controlled, not chaotic)

You want variations that are safely different—hooks, camera moves, settings, wardrobe/colorways—while keeping the rest locked.

3) Guardrails (MUST/AVOID + continuity)

Your meta-prompt should output:

Non-negotiables (“MUST include product in hand”)
Avoid list (“AVOID shaky handheld if you requested locked-off”)
Continuity notes (wardrobe, props, time-of-day)

4) Audio & Dialogue (when relevant)

Veo’s prompt guidance notes that Veo 3 can generate dialogue and you can either give a topic or specify exact lines. (https://deepmind.google/models/veo/prompt-guide/)

5) Fix List (how to iterate without breaking everything)

Your meta-prompt should suggest targeted repairs: “If the scene feels too busy, reduce background elements; if motion is unclear, specify pan/tilt/dolly,” etc.

Step-by-step: Build your Veo3Gen Meta-Prompt (copy/paste template)

Paste the template below into your LLM. Then, each time you have a new idea, you only fill in the placeholders.

Meta-Prompt Template (copy/paste)

Instruction to the LLM:

You are a Veo3Gen Prompt Builder. Generate Veo-ready shot cards and variations using the exact schema requested.

OUTPUTS REQUIRED (in this order):

PROJECT SUMMARY (2–3 sentences)

GLOBAL GUARDRAILS: MUST / AVOID / CONTINUITY (bullets)

SHOT CARDS (exactly {NUM_SHOTS} shots). Each shot must be exactly 6 lines in this order:

Subject:

Action:

Scene:

Camera:

Light:

Style:

CONTROLLED VARIATIONS (exactly {NUM_VARIATIONS}). Each variation must state “CHANGE ONLY: ___” and then provide the revised shot card line(s) only.

FIX LIST (5 bullets): potential failure modes + the smallest prompt edit to fix each.

RULES:

Keep each shot coherent; do not add new characters/props unless listed in MUST.

Avoid conflicting constraints (e.g., handheld + locked-off). If conflicts exist in the inputs, flag them in FIX LIST.

Use specific camera framing and/or camera motion when asked (e.g., low-angle, pan across the scene). (https://deepmind.google/models/veo/prompt-guide/)

Include lighting detail (e.g., warm even lighting, spotlight) for every shot. (https://deepmind.google/models/veo/prompt-guide/)

Include style/texture descriptors (e.g., film noir, 35mm, worn VHS texture) when STYLE is provided. (https://deepmind.google/models/veo/prompt-guide/)

Make locations concrete and evocative (e.g., “smoky jazz club at night” vs “jazz club”). (https://deepmind.google/models/veo/prompt-guide/)

Make subjects specific (age range, hair, notable features) when relevant. (https://deepmind.google/models/veo/prompt-guide/)

Describe action clearly (what happens on screen). (https://deepmind.google/models/veo/prompt-guide/)

If AUDIO includes dialogue, either provide a topic or exact lines per shot, but keep it short and shootable.

PLACEHOLDERS (fill these in):

STYLE: {STYLE}

SUBJECT: {SUBJECT}

ACTION: {ACTION}

SETTING: {SETTING}

CAMERA: {CAMERA}

LIGHT: {LIGHT}

DURATION: {DURATION}

ASPECT: {ASPECT}

AUDIO: {AUDIO}

MUST: {MUST}

AVOID: {AVOID}

PLATFORM/DELIVERABLE: {PLATFORM}

NUM_SHOTS: {NUM_SHOTS}

NUM_VARIATIONS: {NUM_VARIATIONS}

Now generate outputs. Make the shot cards suitable to paste into Veo3Gen with minimal editing.

How to feed the meta-prompt: inputs to paste (so it outputs reliably)

Think of your inputs as a “creative brief” that’s short but specific.

The minimum viable input set

STYLE: The visual identity (e.g., “clean UGC ad, natural skin texture, subtle film grain”)—Veo guidance explicitly encourages describing visual style/texture. (https://deepmind.google/models/veo/prompt-guide/)
SUBJECT: Who/what we see, with concrete detail where useful (more specific beats vague). (https://deepmind.google/models/veo/prompt-guide/)
SETTING: Where it happens, with time-of-day and atmosphere (be vivid). (https://deepmind.google/models/veo/prompt-guide/)
ACTION: What happens in the shot (treat it like blocking). (https://deepmind.google/models/veo/prompt-guide/)
CAMERA + LIGHT: Shot framing/motion + lighting instructions. (https://deepmind.google/models/veo/prompt-guide/)
DURATION + ASPECT: Even if you’re not 100% sure, state your target.
AUDIO: “No dialogue, ambient only” or “short line to camera.” Veo guidance notes dialogue can be generated and can be topic-based or scripted. (https://deepmind.google/models/veo/prompt-guide/)

Optional: use a layered mental model

One useful prompt structure is a layered formula: Camera & Lens + Subject + Action & Physics + Environment + Lighting + Style & Texture + Audio. (https://invideo.io/blog/google-veo-prompt-guide/)

You don’t have to copy that format into your final shot cards—just use it to sanity-check whether you forgot something important.

Example run: one creator idea → 6 shot cards + 12 variations

Scenario: a UGC-style ad for a creator selling an insulated bottle.

Inputs to paste into the meta-prompt (example):

STYLE: UGC ad, crisp natural daylight, subtle film grain, clean color
SUBJECT: creator in late 20s, wavy brown hair, light freckles, casual streetwear
ACTION: show the bottle, demonstrate cold retention with ice, quick lifestyle montage, end with direct-to-camera recommendation
SETTING: small bright kitchen + quick outdoor cutaway near a sunny park
CAMERA: mostly locked-off tripod with one slow push-in for the hook; one close-up product shot
LIGHT: warm even window light in kitchen; outdoor sun with soft fill
DURATION: 15 seconds
ASPECT: 9:16
AUDIO: light upbeat bed + one short spoken line in final shot
MUST: bottle visible in every shot; condensation/ice detail in close-up; end card text: “Cold all day”
AVOID: shaky handheld; cluttered background; unreadable label
PLATFORM: TikTok/Reels
NUM_SHOTS: 6
NUM_VARIATIONS: 12

What “good” output looks like:

Shot 1 hooks fast with a clear camera instruction (e.g., low-angle close framing or slow push-in). Veo guidance recommends specifying framing/motion like low-angle or panning. (https://deepmind.google/models/veo/prompt-guide/)
Shots include concrete lighting direction (warm even window light, etc.). (https://deepmind.google/models/veo/prompt-guide/)
Location is described as more than “kitchen”—texture/atmosphere included. (https://deepmind.google/models/veo/prompt-guide/)
Style is explicit (UGC vibe + subtle texture). (https://deepmind.google/models/veo/prompt-guide/)

Iteration loop: how to request “fix only one variable”

The easiest way to keep consistency is to freeze five lines and edit one.

Controlled variation prompts you can reuse

Ask your LLM:

Change ONLY hook: “For Shot 1, CHANGE ONLY: Action. Keep Subject/Scene/Camera/Light/Style identical.”
Change ONLY camera move: “For Shot 3, CHANGE ONLY: Camera. Provide three options: static, slow pan, slow push-in.”
Change ONLY setting: “For Shot 2, CHANGE ONLY: Scene. Keep the kitchen vibe but move to a ‘smoky jazz club at night’ style equivalent—same level of specificity.” (Location specificity is explicitly recommended.) (https://deepmind.google/models/veo/prompt-guide/)
Change ONLY wardrobe/colorway: “For Shots 1–6, CHANGE ONLY: Subject wardrobe colors (3 palettes). Do not change identity, age, or features.”

This “one-variable” approach is especially helpful when you’re testing multiple creative angles without losing the baseline.

Common failure modes + exact fixes

LLM overwrites your style

Symptom: Your UGC ad suddenly becomes “cinematic film noir.”

Fix: Move STYLE into MUST and add “Do not introduce new genres.” Veo guidance supports explicitly describing style/texture to steer visuals. (https://deepmind.google/models/veo/prompt-guide/)

Vague camera instructions

Symptom: Shots feel generic.

Fix: Require a framing + motion phrase per shot (“close-up, slow push-in,” “low-angle, pan across the scene”). Veo guidance recommends specifying framing/motion. (https://deepmind.google/models/veo/prompt-guide/)

Conflicting constraints

Symptom: “Handheld” appears even though you asked for locked-off.

Fix: Add a conflict detector (see checklist below) and force the LLM to flag conflicts in the FIX LIST.

Mini checklist: quick conflict detection (before you render)

Handheld vs locked-off/tripod both requested
Slow motion paired with real-time dialogue expectations
Macro close-up requested but also “wide establishing” in the same shot
Dark/moody lighting paired with “bright even daylight” in the same shot
Multiple locations crammed into one shot duration

Save it as a system: naming, versioning, weekly refresh

Treat your meta-prompt like a creative asset:

Name it: veo31_meta_prompt_v1_ugc
Version it: increment when you change schema or rules (v1.1, v1.2)
Weekly 2-minute refresh: update your MUST/AVOID based on what broke last week (e.g., “labels unreadable,” “background too busy,” “camera move inconsistent”).

If you later adopt features like “first frame, last frame” for narrative control, keep them as optional inputs so your base template stays stable. (https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1)

Veo 3.1 “Meta-Prompt” Builder for Creators: One Reusable Prompt That Generates Shot Cards, Variations, and Fixes (as of 2026-04-16)